File size: 4,956 Bytes
b84549f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 |
from typing import List
from data.dataloader import build_dataloader
# from methods.elasticdnn.api.online_model import ElasticDNN_OnlineModel
from methods.elasticdnn.api.online_model_v2 import ElasticDNN_OnlineModel
import torch
import sys
from torch import nn
from methods.elasticdnn.api.model import ElasticDNN_OfflineSegFMModel, ElasticDNN_OfflineSegMDModel
from methods.elasticdnn.api.algs.md_pretraining_wo_fbs import ElasticDNN_MDPretrainingWoFBSAlg
from methods.elasticdnn.model.base import ElasticDNNUtil
from methods.elasticdnn.pipeline.offline.fm_to_md.base import FM_to_MD_Util
from methods.elasticdnn.pipeline.offline.fm_to_md.vit import FM_to_MD_ViT_Util
from methods.elasticdnn.pipeline.offline.fm_lora.base import FMLoRA_Util
from methods.elasticdnn.pipeline.offline.fm_lora.vit import FMLoRA_ViT_Util
from methods.elasticdnn.model.vit import ElasticViTUtil
from utils.common.file import ensure_dir
from utils.dl.common.model import LayerActivation, get_module, get_parameter
from utils.common.exp import save_models_dict_for_init, get_res_save_dir
from data import build_scenario
from utils.dl.common.loss import CrossEntropyLossSoft
import torch.nn.functional as F
from utils.dl.common.env import create_tbwriter
import os
from utils.common.log import logger
from utils.common.data_record import write_json
# from methods.shot.shot import OnlineShotModel
from methods.feat_align.main import OnlineFeatAlignModel, FeatAlignAlg
import tqdm
from methods.feat_align.mmd import mmd_rbf
from experiments.utils.elasticfm_da import init_online_model, elasticfm_da
device = 'cuda'
app_name = 'pos'
sd_sparsity = 0.8
settings = {
'involve_fm': True
}
scenario = build_scenario(
source_datasets_name=[i + '-TokenCls' for i in ['HL5Domains-ApexAD2600Progressive', 'HL5Domains-CanonG3', 'HL5Domains-CreativeLabsNomadJukeboxZenXtra40GB',
'HL5Domains-NikonCoolpix4300', 'HL5Domains-Nokia6610']],
target_datasets_order=[i + '-TokenCls' for i in ['Liu3Domains-Computer', 'Liu3Domains-Router', 'Liu3Domains-Speaker',
'Ding9Domains-DiaperChamp', 'Ding9Domains-Norton', 'Ding9Domains-LinksysRouter',
'Ding9Domains-MicroMP3', 'Ding9Domains-Nokia6600', 'Ding9Domains-CanonPowerShotSD500',
'Ding9Domains-ipod', 'Ding9Domains-HitachiRouter', 'Ding9Domains-CanonS100',
'SemEval-Laptop', 'SemEval-Rest'] * 2 + ['Liu3Domains-Computer', 'Liu3Domains-Router']],
da_mode='close_set',
data_dirs={
**{k: f'/data/zql/datasets/nlp_asc_19_domains/dat/absa/Bing5Domains/asc/{k.split("-")[1]}'
for k in [i + '-TokenCls' for i in ['HL5Domains-ApexAD2600Progressive', 'HL5Domains-CanonG3', 'HL5Domains-CreativeLabsNomadJukeboxZenXtra40GB',
'HL5Domains-NikonCoolpix4300', 'HL5Domains-Nokia6610']]},
**{k: f'/data/zql/datasets/nlp_asc_19_domains/dat/absa/Bing3Domains/asc/{k.split("-")[1]}'
for k in [i + '-TokenCls' for i in ['Liu3Domains-Computer', 'Liu3Domains-Router', 'Liu3Domains-Speaker']]},
**{k: f'/data/zql/datasets/nlp_asc_19_domains/dat/absa/Bing9Domains/asc/{k.split("-")[1]}'
for k in [i + '-TokenCls' for i in ['Ding9Domains-DiaperChamp', 'Ding9Domains-Norton', 'Ding9Domains-LinksysRouter',
'Ding9Domains-MicroMP3', 'Ding9Domains-Nokia6600', 'Ding9Domains-CanonPowerShotSD500',
'Ding9Domains-ipod', 'Ding9Domains-HitachiRouter', 'Ding9Domains-CanonS100']]},
**{k: f'/data/zql/datasets/nlp_asc_19_domains/dat/absa/XuSemEval/asc/14/{k.split("-")[1].lower()}'
for k in [i + '-TokenCls' for i in ['SemEval-Laptop', 'SemEval-Rest']]},
},
)
from experiments.elasticdnn.bert_base.online.pos.model import ElasticDNN_POSOnlineModel
elasticfm_model = ElasticDNN_POSOnlineModel('pos', init_online_model(
'experiments/elasticdnn/bert_base/offline/fm_to_md/pos/results/pos_md_w_fbs_index.py/20230704/999998-085253-trial/models/fm_best.pt',
'experiments/elasticdnn/bert_base/offline/fm_to_md/pos/results/pos_md_w_fbs_index.py/20230704/999998-085253-trial/models/md_best.pt',
'pos', __file__
), device, {
'md_to_fm_alpha': 0.1,
'fm_to_md_alpha': 0.1
})
da_alg = FeatAlignAlg
from experiments.elasticdnn.bert_base.online.pos.model import POSOnlineFeatAlignModel
da_model = POSOnlineFeatAlignModel
da_alg_hyp = {
'train_batch_size': 16,
'val_batch_size': 64,
'num_workers': 8,
'optimizer': 'SGD',
'optimizer_args': {'lr': 1e-4, 'momentum': 0.9},
'scheduler': '',
'scheduler_args': {},
'num_iters': 100,
'val_freq': 20,
'feat_align_loss_weight': 1.0,
'sd_sparsity': 0.7
}
elasticfm_da(
[app_name],
[scenario],
[elasticfm_model],
[da_alg],
[da_alg_hyp],
[da_model],
device,
settings,
__file__,
sys.argv[1]
)
|